PERIOD DOMAIN ANALYSIS IN FETAL PULSE OXIMETRY
J. L. Reuss, Ph.D., Member IEEE1, D. E. Bahr, P.E., Senior Member IEEE 2
1
OB Scientific, Inc., Germantown, WI, USA
2
Bahr Management, Middleton, WI, USA
Abstract-The photoplethysmographic signals acquired during
pulse oximetry can be compromised in many ways. Intrapar-
tum fetal pulse oximetry in particular presents challenges to
signal processing. Period domain analysis can overcome the
low pulsatile amplitudes, noise, and maternal modulation
found in these signals. The efficiency of an incremental algo-
rithm reduces the processing requirements for period domain
analysis, facilitating use in low-power and portable devices.
Keywords - Fetal pulse oximetry, spectral analysis, period do-
main analysis, ensemble averaging
I. INTRODUCTION
Pulse oximetry is the non-invasive measurement of
oxygen saturation (SpO2) based upon the relative absorb-
ance of multiple light wavelengths by different species of
hemoglobin. The last decade has seen a renewed interest in
overcoming the limitations of low perfusion and motion in
pulse oximetry, and extending its use into new areas such as
intrapartum fetal monitoring [1].
In fetal pulse oximetry (FPO), the sensor is placed
through the birth canal onto a part of the fetus. Fetal physi-
ology normally operates at a much broader and lower oxy-
gen saturation range, typically SpO2 = 40%-75%, with clini-
cally significant desaturations occurring below 30% [2].
Stability of tissue conditions at the monitoring site, such as
blood volume, become more critical at low saturations, and
calibration of these devices is a challenge [3].
Commercially available devices operate in reflectance
(backscattering) mode, and are placed trans-cervically upon
the fetal body in utero [4]. The pulsatility of the pho-
toplethysmographic (PPG) signals is small compared to
adults or neonates. The intrauterine placement makes am-
bient light interference unlikely, but introduces the possibil-
ity of maternal modulation of the fetal signals. The intimate
proximity of the fetus to maternally vascularized tissue can
result in a strong modulation at the maternal pulse rate.
Traditional time domain techniques employed to proc-
ess PPG signals include peak detection and fiducial point
determination for cardiac period calculation, and peak-
valley measurement for the pulsatile amplitude measure-
ment used in SpO2 calculation. Recently, frequency domain
analysis has been used into algorithms for determination of
the fundamental cardiac frequency and, to some extent, se-
lective removal of noise components based upon frequency
content [5]. These algorithms work in conjunction with
time domain techniques, rather than replacing them.
We propose processing the PPG signals in the period
domain, i.e., determining the relative contributions of differ-
ent periods to the signal content [6]. The advantages of this
method are improved resolution for low frequency biomedi-
cal signals, and compatibility with time domain algorithms.
II. ALGORITHM IMPLEMENTATION
The Discrete Fourier Transform (DFT) and incremental
or “sliding” DFT are fundamental algorithms [7]. For sam-
pling frequency fs the frequency “bin” k of the N-point DFT
corresponds to frequency fk = k · fs / N Hz, and
N −1
X i (k ) = ∑ x(i + n) e − j 2πk n / N k = 0, 1, … N-1 (1)
n =0
is the expression for spectrum of the kth frequency “bin” of
for the sample sequence xi…xi+N-1. At i+1, the “sliding” or
incremental DFT is calculated as
[
X i +1 (k ) = e j 2πk / N X i (k ) + x (i + N ) − x(i ) ] (2)
To derive the Discrete Period Transform (DPT), let s =
1,2,…,N-1 samples be the range of periods possible in the
sequence xi…xi+N-1. Frequency fk corresponds to period sk =
1/fk = N / ( k · fs) seconds = N / k samples, so k = N / sk . Sub-
stituting into (1) and (2), for the period s,
N −1
T i (s) = ∑ x(i + n) e − j 2πn / s
s= 1, 2, …, N-1 (3)
n =0
and
[ ]
T i +1 (s) = e j 2π / s T i (s) + x(i + N ) − x(i) (4)
The DPT calculates the period spectrum at a resolution
1/fs. Over the relatively small frequency range of PPG sig-
nals (approx. 0.1-10Hz), this resolution is achieved with
modest processing power and memory. No conversion from
frequency is necessary for compatibility with time domain
algorithms, an advantage where period measurements are
interchanged between power spectrum and time domains.
Ensemble averaging has been applied to PPG signals,
usually employing an external cardiac “trigger” obtained
from an ECG source [8]. This has also been attempted in
fetal pulse oximetry, although obtaining a reliable fetal ECG
signal generally requires use of a fetal-invasive scalp elec-
trode. By deriving a reliable cardiac period estimate from
period domain analysis, ensemble averaging may proceed
without an external trigger source.
III. METHODOLOGY
A fetal signal database was collected to investigate
pulse oximetry algorithms in a personal-computer-based
Fetal Oximetry Platform (FOP). The OB Scientific™ OBS-
900 Fetal Oxygen Sensor was placed through the birth canal
onto the fetal torso. Red, infrared, and dark signals from the
sensor, digitized at a rate of 120 samples/second and 21 bit
resolution, were transmitted from the OBS-500 Fetal Pulse
Oximeter to a portable computer for storage. The FOP sys-
tem incorporates the oximeter’s algorithms as well as a user
interface for graphical and textual display and analysis of
the oximeter operation. FOP permits a signal record to be
re-run with alterations to the algorithms and subsequent
pulse-to-pulse comparison of the results.
IV. RESULTS
A. Pulse Rate Determination
Fig. 1 illustrates an example of weak signal data (red
and infrared (IR)) in the time domain (a-c), and the corre-
sponding period domain spectrum (d). The peak in (d) is
the cardiac period (fetal heart rate 137 beats/minute).
Red
(a)
IR
Red
(b)
IR 1.0
period (sec)
Red
(c)
IR
0
time (sec) 5 the halved rate will still be effective, if the temporal rela-
Fig. 1. Weak photoplethysmographic signals: (a) unprocessed,
(b) band-passed, (c) ensemble averaged, (d) period domain
Fig. 1 (a) is the unfiltered signal data from the sensor.
The result of simple band-pass filtering to obtain the pulsa-
tile portion of each signal is shown in (b). Reliable deter-
mination of the pulse rate from (b) would be difficult. The
cardiac period is apparent in (d). The fetal signal compo-
nents extracted using ensemble averaging with the cardiac
period estimate of (d) are shown in (c). The period domain
spectrum yields a useful cardiac period for rate calculation
that can be tracked through all but extreme, prolonged noise.
B. Maternal modulation
Fig. 2 illustrates a case of significant maternal modula-
tion in the time domain (a-c) and the period domain (d).
The right peak in (d) is the cardiac period (fetal heart rate
126 beats/minute), whereas the left peak is the maternal
cardiac period (97 beats/minute heart rate).
Red
(a)
IR
Red
(b)
IR 1.0 0.5 0.25
period (sec)
Red
(c)
IR
0 5
time (sec)
Fig. 2. Maternal modulation in PPG signals: (a) unprocessed,
(b) band-passed, (c) ensemble averaged, (d) period domain
The result of simple band-pass filtering to obtain the
pulsatile portion of each signal is shown in Fig. 2 (b). En-
semble averaging utilizing the cardiac period derived from
period domain analysis can better extract the fetal signal
components, as illustrated in (c). Errors in pulse rate and
oxygen saturation calculations could result from processing
the signals in (b). Applying ensemble averaging results in a
much cleaner signal (c) for further time domain analysis.
V. DISCUSSION
Biomedical signals are notorious for lacking statistical
stationarity. The PPG period domain spectrum obtained
with the incremental DPT algorithm tracks relatively fast
heart rate changes (such as heart rate accelerations and de-
celerations), but may not produce useful results in the pres-
ence of arrhythmias. Because it is used in conjunction with
time domain pulse detection, the period domain algorithm
may be disabled automatically in such circumstances.
In the case of fetal pulse oximetry, highly irregular
0.5 0.25 rhythms are relatively uncommon. When bigeminy is pre-
sent with regularity of pulse spacing, period tracking may
“lock on” and track the normal (hemodynamically strong)
(d) pulses at half the actual pulse rate. Ensemble averaging at
tionship of normal and PVC beats is consistent.
VI. CONCLUSION
Period domain analysis utilizing an incremental DPT
algorithm is an effective and efficient way to process peri-
odic biomedical signals for spectral content. It provides the
capabilities of frequency domain analysis, with certain ad-
vantages in implementation. Processing of photoplethys-
mographic fetal pulse oximeter signals with period domain
analysis improves pulse rate availability and accuracy, and
permits removal of interference by techniques such as en-
semble averaging without an external noise reference.
REFERENCES
[1] C. D. Kurth, J. M. Steven, D. Swedlow, “New frontiers in oximetry,”
Am J Anesthesiol vol 23, pp. 169-175, 1996.
[2] G.A. Dildy, J.A. Thorp, J.D. Yeast and S.L. Clark, "The relationship
between oxygen saturation and pH in umbilical blood: implications for
intrapartum fetal oxygen saturation monitoring," Am J Obstet Gynecol vol.
175, no. 3 Pt 1, pp. 682-687, 1996.
[3] J. L. Reuss and D. Siker, “Monitoring site and wavelength selection for
fetal pulse oximetry,” Assoc. Advanc. Med. Instr. Annual Meeting, June
2001.
[4] A.K. Luttkus, J.H. Stupin, M. Porath and J.W. Dudenhausen, "Evalua-
tion of signal quality of a new fetal pulsoximetry system (OBS-500). A
recent development of a fetal oxisensor positioned on the fetal back," J
Perinat Med vol. 29, no. S1, pp. 251, 2001.
[5] T.L. Rusch, R. Sankar and J.E. Scharf, "Signal processing methods for
pulse oximetry," Comput Biol Med vol. 26, no. 2, pp. 143-159, 1996.
[6] D. E. Bahr and J. L. Reuss, “Method and Apparatus for Processing a
(d) Physiological Signal,” U. S. Patent 6,339,715 B1, January 15, 2002.
[7] L. R. Rabiner and B. Gold, Theory and Application of Digital Signal
Processing. Englewood Cliffs, NJ: Prentice-Hall, 1975.
[8] S. Palreddy, “Signal Processing Algorithms,” in Design of Pulse
Oximeters, J. G. Webster, Ed. Bristol and Philadelphia: Institute of Physics
Publishing, 1997, pp. 124-158.